Most apps don't need a chatbot
AI belongs where it removes work, not where it performs intelligence. A short field guide to telling the difference.
Every second brief we see now has the same line in it: 'and AI features.' Fair enough — the pressure is real, investors ask about it, users half-expect it. But 'add AI' is not a feature request. It's a mood. And the default translation of that mood — a chat window bolted to the corner of an app that already had a job to do — is usually the wrong one.
We've shipped conversational AI where conversation was the whole point of the product. That's the exception. When chat is the product, chat is right. When chat is an accessory, it's a burden wearing a glow-up.
The test: does it remove steps?
Good AI integration is mostly invisible. It turns a photo into a filled-in form. It drafts the message the user was about to write anyway. It files, sorts, and flags so a human looks at ten things instead of two hundred. In every case the user does less than before. A bolted-on chatbot fails this test: now the user has to compose a prompt, read an essay back, and judge whether to trust it. That's more work, not less — outsourced to your busiest resource.
Where it earns its place
- Summarizing what happened while the user was away, instead of a feed they'll never scroll through.
- Turning messy input — photos, voice notes, pasted text — into structured data without a form.
- Writing the first draft of anything people write repeatedly and reluctantly.
- Ranking and flagging, so attention lands where it matters.
Notice that none of these require the user to know AI is involved. The best compliment an AI feature can get is nobody mentioning it — things just take fewer steps than they used to.
The cost of pretending
Every AI feature is a promise with a meter running: latency the user feels, tokens someone pays for, and a new surface where the product can be confidently wrong. That's a fine price for removed work. It's a terrible price for a demo. One AI feature that genuinely saves people time will outlast five that impressed a boardroom.
So the roadmap question isn't 'where can we put AI.' It's 'what work can we make disappear.' Answer that, and the technology chooses itself.